Detail publikace

Optimization of Power Analysis Using Neural Network

MARTINÁSEK, Z. HAJNÝ, J. MALINA, L.

Originální název

Optimization of Power Analysis Using Neural Network

Anglický název

Optimization of Power Analysis Using Neural Network

Jazyk

en

Originální abstrakt

In power analysis, many different statistical methods and power consumption models are used to obtain the value of a secret key from the power traces measured. An interesting method of power analysis based on multi-layer perceptron claiming a $90\%$ success rate. The theoretical and empirical success rates were determined to be $80\%$ and $85\%$, respectively, which is not sufficient enough. In the paper, we propose and realize an optimization of this power analysis method which improves the success rate to almost $100\%$. The optimization is based on preprocessing the measured power traces using the calculation of the average trace and the subsequent calculation of the difference power traces. In this way, the prepared power patterns were used for neural network training and of course during the attack. This optimization is computationally undemanding compared to other methods of preprocessing usually applied in power analysis, and has a great impact on classification results. In the paper, we compare the results of the optimized method with the original implementation. We highlight positive and also some negative impacts of the optimization on classification results.

Anglický abstrakt

In power analysis, many different statistical methods and power consumption models are used to obtain the value of a secret key from the power traces measured. An interesting method of power analysis based on multi-layer perceptron claiming a $90\%$ success rate. The theoretical and empirical success rates were determined to be $80\%$ and $85\%$, respectively, which is not sufficient enough. In the paper, we propose and realize an optimization of this power analysis method which improves the success rate to almost $100\%$. The optimization is based on preprocessing the measured power traces using the calculation of the average trace and the subsequent calculation of the difference power traces. In this way, the prepared power patterns were used for neural network training and of course during the attack. This optimization is computationally undemanding compared to other methods of preprocessing usually applied in power analysis, and has a great impact on classification results. In the paper, we compare the results of the optimized method with the original implementation. We highlight positive and also some negative impacts of the optimization on classification results.

Dokumenty

BibTex


@inproceedings{BUT108335,
  author="Zdeněk {Martinásek} and Jan {Hajný} and Lukáš {Malina}",
  title="Optimization of Power Analysis Using Neural Network",
  annote="In power analysis, many different statistical methods and power consumption models are used to obtain the value of a secret key from the power traces measured. An interesting method of power analysis based on multi-layer perceptron claiming a $90\%$ success rate. The theoretical and empirical success rates were determined to be $80\%$ and $85\%$, respectively, which is not sufficient enough. In the paper, we propose and realize an optimization of this power analysis method which improves the success rate to almost $100\%$. The optimization is based on preprocessing the measured power traces using the calculation of the average trace and the subsequent calculation of the difference power traces. In this way, the prepared power patterns were used for neural network training and of course during the attack. This optimization is computationally undemanding compared to other methods of preprocessing usually applied in power analysis, and has a great impact on classification results. In the paper, we compare the results of the optimized method with the original implementation. We highlight positive and also some negative impacts of the optimization on classification results.",
  address="Springer",
  booktitle="Smart Card Research and Advanced Applications, Lecture Notes in Computer Science",
  chapter="108335",
  doi="10.1007/978-3-319-08302-5_7",
  howpublished="online",
  institution="Springer",
  year="2014",
  month="july",
  pages="94--107",
  publisher="Springer",
  type="conference paper"
}